dc.description.abstract | The major objective of this study was to determine the predictability of East
African seasonal rainfall using sea surface temperatures as the major predictors. Due
to the problems associated with timely availability of SST records, the second
objective attempted to search for some potential SST proxies which could be used
as alternative predictors. Under this concept SST proxies were searched from the
surface air temperatures of some coastal/island stations, as well as 200 Hectapascal
(Hpa) temperatures. The area of the study namely East African region is located
within latitudes 5°N and 12°S and longitudes 29°E and 43°E.
The seasonal SST data which were used in this study were generated from
SST grid point data which were obtained from United Kingdom Meteorological Office
(UK. Met.office). Point rainfall records were obtained through the respective
meteorological departments. The rainfall records were quality controlled before they
were used in the study. The SST data sets used had already been subjected to various
quality control methods by the UK. Met. office. The quality controlled data were
subjected to the various analyses including studies of space-time rainfall anomaly
patterns, trends, spectral, correlation, principal component and regression analyses.
Both zero and time lagged correlation analyses were used to search for the potential
predictors. Principal component analysis was however used first to delineate the
complex relationships between various variables and predictors. The final part of the
study used step wise regression methods to fit the best regression equations between
the predictand (rainfall) and the various predictors.
The period 1940-1975 was used for model building while the period
1976-1990 was used for testing the skill of the developed models. The results from the
study indicated that dry and wet anomaly patterns were recurrent in the seasonal
rainfall time series. There were however significant Spatial variations in the observed
anomaly patterns with exceptions of some years in which the spatial anomaly patterns
were relatively similar. These included some of the El-Niiio and la-Nina years. Trend
analysis detected no significant changes in the observed inter annual rainfall patterns
aaaat most locations.
Spectral analysis however delineated some dominant spectral bands of about
2.2-2.8, 3.0-3.7,4.9-6.0, and 10-12.5 years. The first three peaks were detectable
in all the seasons and at all locations. There were however large spatial variations in
the variances accounted for by the various spectral peaks at the individual locations.
The 10-12.5 year peak was dominant in the shores of lake Victoria as well as regions
of high terrain. The 2.2-2.8 year cycle has been associated with Quasi-biennial
oscillation (QBO) while the 3.0-3.7 and 5-6 cycles are common in the El-nifio and
southern oscillation temporal patterns. Both systems have been observed to be
significantly correlated with rainfall over parts of East Africa. The 10-12.5 years
cycle has been associated with sunspot cycles by a number of authors.
The results from Principal component analysis (PCA) highlighted the
significant differences in the seasonal rainfall characteristics. The number of
significant PCA modes and variance accounted for by SST varied significantly from
season to season. Maximum and minimum seasonal variances were however
accounted for by SST during the September-November and March-May rainfall
seasons respectively.
Correlation analysis indicated significant correlation between regional rainfall
and SST over some specific ocean regions. These formed the fundamental base for
the predictors which were used in this study. The results from the study further
showed that surface temperatures at some coastal/island stations namely, Larnu and
Mombasal Aldabra and Seychelles respectively were closely correlated with some of
the SST modes which had correlated significantly with rainfall. This signifies the
likely use of some of the coastall island surface temperatures as proxies of some of
the SST predictors. Low correlation was however observed with the 200 Hpa
temperature.
Results from regression analysis indicated that although SST had some skill
In the prediction of seasonal rainfall during some seasons, seasonal variance
accounted by SST were generally less than 50% with exception of June-August and
September-November seasons when relatively high percentages of 71.6% and 58.7%
respectively were observed. Time coefficients of the SSTs explained a maximum
variance of the seasonal rainfall of 59.5%,54.1 %,57.5%,49.2% for the seasons of
December to February, March to May, June to August and September to November
respectively .
In general, higher prediction skills were recorded during the years with larger
SST anomaly signals like in the cases of la-Nina and El-niiio years. The results from
this study indicate that SST modes could be used to give some useful lead time
seasonal rainfall information which could help to minimize the severe negative socioeconomic
impacts of wide spread rainfall anomalies which are common in the region.
Such information are not only useful in the planning and management of rain dependant
activities but can also form a crucial component of any early warning
system for shortage of food, water, energy, and other basic socio-economic activities | en |